Hacker News Reader

Hacker News

Comment by Gwaih1r | Hacker News Reader

Full thread

Gwaih1r

I had a quick Google and couldn't find any published benchmarks yet. Does anyone know of one so far?

View on HN

According to a German news article [0], Intel published benchmarks themselves [1].

[0]: https://heise.de/-4144515

[1]: https://www.intel.com/content/www/us/en/architecture-and-tec...

Can anyone explain these graphs a bit. I.e the first one shows no impact what-so-ever, but are the benchmark conditions real-life-y enough?

Well, the first two aren't even interesting at all, because the fix is not needed there and it is disabled.

The first two just show, that the fix doesn't make performance worse when disabled.

"With non-virtualized operating systems."

"For non-virtualized environments or where it can be guaranteed that all virtualized guest operating systems are trusted." --> Aka disabled.

Intel's attempt to suppress independent benchmarking suggests that this is not the whole story. This suspicion lingers even though it has removed the clause.

There exists other reasons to not allow third-party benchmarking, other than trying to hide something. A common reason is as a way to make sure the benchmark is done in a proper way.

I've had the experience myself, and I can imagine that when a system is complex and hard to measure that you don't want an internet journalist running Cinebench to become the reference benchmark for your customers.

That's a fair point, though the best solution would be for Intel to commission independent or independently-audited benchmark tests.

Yeah, but what if the microcode killed cinebench performance? Wouldn't that be interesting?

The problem is still the context in which cinebench would be ran. There are so many variable on a complex system that "the microcode killed cinebench perf" could very well be an over simplification of "the microcode in conjunction with this particular motherboard and this version of the kernel plus this particular Nvidia driver lead to a bad interaction that affects cinebench performance negatively".

This is all I've seen so far.

https://www.phoronix.com/scan.php?page=article&item=l1tf-for...

While interesting, those benchmarks are only for the kernel mitigations; they don't benchmark the new microcode:

> Update: To note, no microcode changes/updates were made to the systems under test for this article, just testing/comparing the kernel patches.

Reply on news.ycombinator.com